4 * Kernel internal timers, basic process system calls
6 * Copyright (C) 1991, 1992 Linus Torvalds
8 * 1997-01-28 Modified by Finn Arne Gangstad to make timers scale better.
10 * 1997-09-10 Updated NTP code according to technical memorandum Jan '96
11 * "A Kernel Model for Precision Timekeeping" by Dave Mills
12 * 1998-12-24 Fixed a xtime SMP race (we need the xtime_lock rw spinlock to
13 * serialize accesses to xtime/lost_ticks).
14 * Copyright (C) 1998 Andrea Arcangeli
15 * 1999-03-10 Improved NTP compatibility by Ulrich Windl
16 * 2002-05-31 Move sys_sysinfo here and make its locking sane, Robert Love
17 * 2000-10-05 Implemented scalable SMP per-CPU timer handling.
18 * Copyright (C) 2000, 2001, 2002 Ingo Molnar
19 * Designed by David S. Miller, Alexey Kuznetsov and Ingo Molnar
22 #include <linux/kernel_stat.h>
23 #include <linux/module.h>
24 #include <linux/interrupt.h>
25 #include <linux/percpu.h>
26 #include <linux/init.h>
28 #include <linux/swap.h>
29 #include <linux/pid_namespace.h>
30 #include <linux/notifier.h>
31 #include <linux/thread_info.h>
32 #include <linux/time.h>
33 #include <linux/jiffies.h>
34 #include <linux/posix-timers.h>
35 #include <linux/cpu.h>
36 #include <linux/syscalls.h>
37 #include <linux/delay.h>
38 #include <linux/tick.h>
39 #include <linux/kallsyms.h>
40 #include <linux/perf_event.h>
41 #include <linux/sched.h>
42 #include <linux/slab.h>
44 #include <asm/uaccess.h>
45 #include <asm/unistd.h>
46 #include <asm/div64.h>
47 #include <asm/timex.h>
50 #define CREATE_TRACE_POINTS
51 #include <trace/events/timer.h>
53 u64 jiffies_64 __cacheline_aligned_in_smp
= INITIAL_JIFFIES
;
55 EXPORT_SYMBOL(jiffies_64
);
58 * per-CPU timer vector definitions:
60 #define TVN_BITS (CONFIG_BASE_SMALL ? 4 : 6)
61 #define TVR_BITS (CONFIG_BASE_SMALL ? 6 : 8)
62 #define TVN_SIZE (1 << TVN_BITS)
63 #define TVR_SIZE (1 << TVR_BITS)
64 #define TVN_MASK (TVN_SIZE - 1)
65 #define TVR_MASK (TVR_SIZE - 1)
68 struct list_head vec
[TVN_SIZE
];
72 struct list_head vec
[TVR_SIZE
];
77 struct timer_list
*running_timer
;
78 unsigned long timer_jiffies
;
79 unsigned long next_timer
;
85 } ____cacheline_aligned
;
87 struct tvec_base boot_tvec_bases
;
88 EXPORT_SYMBOL(boot_tvec_bases
);
89 static DEFINE_PER_CPU(struct tvec_base
*, tvec_bases
) = &boot_tvec_bases
;
92 * Note that all tvec_bases are 2 byte aligned and lower bit of
93 * base in timer_list is guaranteed to be zero. Use the LSB for
94 * the new flag to indicate whether the timer is deferrable
96 #define TBASE_DEFERRABLE_FLAG (0x1)
98 /* Functions below help us manage 'deferrable' flag */
99 static inline unsigned int tbase_get_deferrable(struct tvec_base
*base
)
101 return ((unsigned int)(unsigned long)base
& TBASE_DEFERRABLE_FLAG
);
104 static inline struct tvec_base
*tbase_get_base(struct tvec_base
*base
)
106 return ((struct tvec_base
*)((unsigned long)base
& ~TBASE_DEFERRABLE_FLAG
));
109 static inline void timer_set_deferrable(struct timer_list
*timer
)
111 timer
->base
= ((struct tvec_base
*)((unsigned long)(timer
->base
) |
112 TBASE_DEFERRABLE_FLAG
));
116 timer_set_base(struct timer_list
*timer
, struct tvec_base
*new_base
)
118 timer
->base
= (struct tvec_base
*)((unsigned long)(new_base
) |
119 tbase_get_deferrable(timer
->base
));
122 static unsigned long round_jiffies_common(unsigned long j
, int cpu
,
126 unsigned long original
= j
;
129 * We don't want all cpus firing their timers at once hitting the
130 * same lock or cachelines, so we skew each extra cpu with an extra
131 * 3 jiffies. This 3 jiffies came originally from the mm/ code which
133 * The skew is done by adding 3*cpunr, then round, then subtract this
134 * extra offset again.
141 * If the target jiffie is just after a whole second (which can happen
142 * due to delays of the timer irq, long irq off times etc etc) then
143 * we should round down to the whole second, not up. Use 1/4th second
144 * as cutoff for this rounding as an extreme upper bound for this.
145 * But never round down if @force_up is set.
147 if (rem
< HZ
/4 && !force_up
) /* round down */
152 /* now that we have rounded, subtract the extra skew again */
155 if (j
<= jiffies
) /* rounding ate our timeout entirely; */
161 * __round_jiffies - function to round jiffies to a full second
162 * @j: the time in (absolute) jiffies that should be rounded
163 * @cpu: the processor number on which the timeout will happen
165 * __round_jiffies() rounds an absolute time in the future (in jiffies)
166 * up or down to (approximately) full seconds. This is useful for timers
167 * for which the exact time they fire does not matter too much, as long as
168 * they fire approximately every X seconds.
170 * By rounding these timers to whole seconds, all such timers will fire
171 * at the same time, rather than at various times spread out. The goal
172 * of this is to have the CPU wake up less, which saves power.
174 * The exact rounding is skewed for each processor to avoid all
175 * processors firing at the exact same time, which could lead
176 * to lock contention or spurious cache line bouncing.
178 * The return value is the rounded version of the @j parameter.
180 unsigned long __round_jiffies(unsigned long j
, int cpu
)
182 return round_jiffies_common(j
, cpu
, false);
184 EXPORT_SYMBOL_GPL(__round_jiffies
);
187 * __round_jiffies_relative - function to round jiffies to a full second
188 * @j: the time in (relative) jiffies that should be rounded
189 * @cpu: the processor number on which the timeout will happen
191 * __round_jiffies_relative() rounds a time delta in the future (in jiffies)
192 * up or down to (approximately) full seconds. This is useful for timers
193 * for which the exact time they fire does not matter too much, as long as
194 * they fire approximately every X seconds.
196 * By rounding these timers to whole seconds, all such timers will fire
197 * at the same time, rather than at various times spread out. The goal
198 * of this is to have the CPU wake up less, which saves power.
200 * The exact rounding is skewed for each processor to avoid all
201 * processors firing at the exact same time, which could lead
202 * to lock contention or spurious cache line bouncing.
204 * The return value is the rounded version of the @j parameter.
206 unsigned long __round_jiffies_relative(unsigned long j
, int cpu
)
208 unsigned long j0
= jiffies
;
210 /* Use j0 because jiffies might change while we run */
211 return round_jiffies_common(j
+ j0
, cpu
, false) - j0
;
213 EXPORT_SYMBOL_GPL(__round_jiffies_relative
);
216 * round_jiffies - function to round jiffies to a full second
217 * @j: the time in (absolute) jiffies that should be rounded
219 * round_jiffies() rounds an absolute time in the future (in jiffies)
220 * up or down to (approximately) full seconds. This is useful for timers
221 * for which the exact time they fire does not matter too much, as long as
222 * they fire approximately every X seconds.
224 * By rounding these timers to whole seconds, all such timers will fire
225 * at the same time, rather than at various times spread out. The goal
226 * of this is to have the CPU wake up less, which saves power.
228 * The return value is the rounded version of the @j parameter.
230 unsigned long round_jiffies(unsigned long j
)
232 return round_jiffies_common(j
, raw_smp_processor_id(), false);
234 EXPORT_SYMBOL_GPL(round_jiffies
);
237 * round_jiffies_relative - function to round jiffies to a full second
238 * @j: the time in (relative) jiffies that should be rounded
240 * round_jiffies_relative() rounds a time delta in the future (in jiffies)
241 * up or down to (approximately) full seconds. This is useful for timers
242 * for which the exact time they fire does not matter too much, as long as
243 * they fire approximately every X seconds.
245 * By rounding these timers to whole seconds, all such timers will fire
246 * at the same time, rather than at various times spread out. The goal
247 * of this is to have the CPU wake up less, which saves power.
249 * The return value is the rounded version of the @j parameter.
251 unsigned long round_jiffies_relative(unsigned long j
)
253 return __round_jiffies_relative(j
, raw_smp_processor_id());
255 EXPORT_SYMBOL_GPL(round_jiffies_relative
);
258 * __round_jiffies_up - function to round jiffies up to a full second
259 * @j: the time in (absolute) jiffies that should be rounded
260 * @cpu: the processor number on which the timeout will happen
262 * This is the same as __round_jiffies() except that it will never
263 * round down. This is useful for timeouts for which the exact time
264 * of firing does not matter too much, as long as they don't fire too
267 unsigned long __round_jiffies_up(unsigned long j
, int cpu
)
269 return round_jiffies_common(j
, cpu
, true);
271 EXPORT_SYMBOL_GPL(__round_jiffies_up
);
274 * __round_jiffies_up_relative - function to round jiffies up to a full second
275 * @j: the time in (relative) jiffies that should be rounded
276 * @cpu: the processor number on which the timeout will happen
278 * This is the same as __round_jiffies_relative() except that it will never
279 * round down. This is useful for timeouts for which the exact time
280 * of firing does not matter too much, as long as they don't fire too
283 unsigned long __round_jiffies_up_relative(unsigned long j
, int cpu
)
285 unsigned long j0
= jiffies
;
287 /* Use j0 because jiffies might change while we run */
288 return round_jiffies_common(j
+ j0
, cpu
, true) - j0
;
290 EXPORT_SYMBOL_GPL(__round_jiffies_up_relative
);
293 * round_jiffies_up - function to round jiffies up to a full second
294 * @j: the time in (absolute) jiffies that should be rounded
296 * This is the same as round_jiffies() except that it will never
297 * round down. This is useful for timeouts for which the exact time
298 * of firing does not matter too much, as long as they don't fire too
301 unsigned long round_jiffies_up(unsigned long j
)
303 return round_jiffies_common(j
, raw_smp_processor_id(), true);
305 EXPORT_SYMBOL_GPL(round_jiffies_up
);
308 * round_jiffies_up_relative - function to round jiffies up to a full second
309 * @j: the time in (relative) jiffies that should be rounded
311 * This is the same as round_jiffies_relative() except that it will never
312 * round down. This is useful for timeouts for which the exact time
313 * of firing does not matter too much, as long as they don't fire too
316 unsigned long round_jiffies_up_relative(unsigned long j
)
318 return __round_jiffies_up_relative(j
, raw_smp_processor_id());
320 EXPORT_SYMBOL_GPL(round_jiffies_up_relative
);
323 * set_timer_slack - set the allowed slack for a timer
324 * @slack_hz: the amount of time (in jiffies) allowed for rounding
326 * Set the amount of time, in jiffies, that a certain timer has
327 * in terms of slack. By setting this value, the timer subsystem
328 * will schedule the actual timer somewhere between
329 * the time mod_timer() asks for, and that time plus the slack.
331 * By setting the slack to -1, a percentage of the delay is used
334 void set_timer_slack(struct timer_list
*timer
, int slack_hz
)
336 timer
->slack
= slack_hz
;
338 EXPORT_SYMBOL_GPL(set_timer_slack
);
341 static inline void set_running_timer(struct tvec_base
*base
,
342 struct timer_list
*timer
)
345 base
->running_timer
= timer
;
349 static void internal_add_timer(struct tvec_base
*base
, struct timer_list
*timer
)
351 unsigned long expires
= timer
->expires
;
352 unsigned long idx
= expires
- base
->timer_jiffies
;
353 struct list_head
*vec
;
355 if (idx
< TVR_SIZE
) {
356 int i
= expires
& TVR_MASK
;
357 vec
= base
->tv1
.vec
+ i
;
358 } else if (idx
< 1 << (TVR_BITS
+ TVN_BITS
)) {
359 int i
= (expires
>> TVR_BITS
) & TVN_MASK
;
360 vec
= base
->tv2
.vec
+ i
;
361 } else if (idx
< 1 << (TVR_BITS
+ 2 * TVN_BITS
)) {
362 int i
= (expires
>> (TVR_BITS
+ TVN_BITS
)) & TVN_MASK
;
363 vec
= base
->tv3
.vec
+ i
;
364 } else if (idx
< 1 << (TVR_BITS
+ 3 * TVN_BITS
)) {
365 int i
= (expires
>> (TVR_BITS
+ 2 * TVN_BITS
)) & TVN_MASK
;
366 vec
= base
->tv4
.vec
+ i
;
367 } else if ((signed long) idx
< 0) {
369 * Can happen if you add a timer with expires == jiffies,
370 * or you set a timer to go off in the past
372 vec
= base
->tv1
.vec
+ (base
->timer_jiffies
& TVR_MASK
);
375 /* If the timeout is larger than 0xffffffff on 64-bit
376 * architectures then we use the maximum timeout:
378 if (idx
> 0xffffffffUL
) {
380 expires
= idx
+ base
->timer_jiffies
;
382 i
= (expires
>> (TVR_BITS
+ 3 * TVN_BITS
)) & TVN_MASK
;
383 vec
= base
->tv5
.vec
+ i
;
388 list_add_tail(&timer
->entry
, vec
);
391 #ifdef CONFIG_TIMER_STATS
392 void __timer_stats_timer_set_start_info(struct timer_list
*timer
, void *addr
)
394 if (timer
->start_site
)
397 timer
->start_site
= addr
;
398 memcpy(timer
->start_comm
, current
->comm
, TASK_COMM_LEN
);
399 timer
->start_pid
= current
->pid
;
402 static void timer_stats_account_timer(struct timer_list
*timer
)
404 unsigned int flag
= 0;
406 if (likely(!timer
->start_site
))
408 if (unlikely(tbase_get_deferrable(timer
->base
)))
409 flag
|= TIMER_STATS_FLAG_DEFERRABLE
;
411 timer_stats_update_stats(timer
, timer
->start_pid
, timer
->start_site
,
412 timer
->function
, timer
->start_comm
, flag
);
416 static void timer_stats_account_timer(struct timer_list
*timer
) {}
419 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
421 static struct debug_obj_descr timer_debug_descr
;
424 * fixup_init is called when:
425 * - an active object is initialized
427 static int timer_fixup_init(void *addr
, enum debug_obj_state state
)
429 struct timer_list
*timer
= addr
;
432 case ODEBUG_STATE_ACTIVE
:
433 del_timer_sync(timer
);
434 debug_object_init(timer
, &timer_debug_descr
);
442 * fixup_activate is called when:
443 * - an active object is activated
444 * - an unknown object is activated (might be a statically initialized object)
446 static int timer_fixup_activate(void *addr
, enum debug_obj_state state
)
448 struct timer_list
*timer
= addr
;
452 case ODEBUG_STATE_NOTAVAILABLE
:
454 * This is not really a fixup. The timer was
455 * statically initialized. We just make sure that it
456 * is tracked in the object tracker.
458 if (timer
->entry
.next
== NULL
&&
459 timer
->entry
.prev
== TIMER_ENTRY_STATIC
) {
460 debug_object_init(timer
, &timer_debug_descr
);
461 debug_object_activate(timer
, &timer_debug_descr
);
468 case ODEBUG_STATE_ACTIVE
:
477 * fixup_free is called when:
478 * - an active object is freed
480 static int timer_fixup_free(void *addr
, enum debug_obj_state state
)
482 struct timer_list
*timer
= addr
;
485 case ODEBUG_STATE_ACTIVE
:
486 del_timer_sync(timer
);
487 debug_object_free(timer
, &timer_debug_descr
);
494 static struct debug_obj_descr timer_debug_descr
= {
495 .name
= "timer_list",
496 .fixup_init
= timer_fixup_init
,
497 .fixup_activate
= timer_fixup_activate
,
498 .fixup_free
= timer_fixup_free
,
501 static inline void debug_timer_init(struct timer_list
*timer
)
503 debug_object_init(timer
, &timer_debug_descr
);
506 static inline void debug_timer_activate(struct timer_list
*timer
)
508 debug_object_activate(timer
, &timer_debug_descr
);
511 static inline void debug_timer_deactivate(struct timer_list
*timer
)
513 debug_object_deactivate(timer
, &timer_debug_descr
);
516 static inline void debug_timer_free(struct timer_list
*timer
)
518 debug_object_free(timer
, &timer_debug_descr
);
521 static void __init_timer(struct timer_list
*timer
,
523 struct lock_class_key
*key
);
525 void init_timer_on_stack_key(struct timer_list
*timer
,
527 struct lock_class_key
*key
)
529 debug_object_init_on_stack(timer
, &timer_debug_descr
);
530 __init_timer(timer
, name
, key
);
532 EXPORT_SYMBOL_GPL(init_timer_on_stack_key
);
534 void destroy_timer_on_stack(struct timer_list
*timer
)
536 debug_object_free(timer
, &timer_debug_descr
);
538 EXPORT_SYMBOL_GPL(destroy_timer_on_stack
);
541 static inline void debug_timer_init(struct timer_list
*timer
) { }
542 static inline void debug_timer_activate(struct timer_list
*timer
) { }
543 static inline void debug_timer_deactivate(struct timer_list
*timer
) { }
546 static inline void debug_init(struct timer_list
*timer
)
548 debug_timer_init(timer
);
549 trace_timer_init(timer
);
553 debug_activate(struct timer_list
*timer
, unsigned long expires
)
555 debug_timer_activate(timer
);
556 trace_timer_start(timer
, expires
);
559 static inline void debug_deactivate(struct timer_list
*timer
)
561 debug_timer_deactivate(timer
);
562 trace_timer_cancel(timer
);
565 static void __init_timer(struct timer_list
*timer
,
567 struct lock_class_key
*key
)
569 timer
->entry
.next
= NULL
;
570 timer
->base
= __raw_get_cpu_var(tvec_bases
);
572 #ifdef CONFIG_TIMER_STATS
573 timer
->start_site
= NULL
;
574 timer
->start_pid
= -1;
575 memset(timer
->start_comm
, 0, TASK_COMM_LEN
);
577 lockdep_init_map(&timer
->lockdep_map
, name
, key
, 0);
580 void setup_deferrable_timer_on_stack_key(struct timer_list
*timer
,
582 struct lock_class_key
*key
,
583 void (*function
)(unsigned long),
586 timer
->function
= function
;
588 init_timer_on_stack_key(timer
, name
, key
);
589 timer_set_deferrable(timer
);
591 EXPORT_SYMBOL_GPL(setup_deferrable_timer_on_stack_key
);
594 * init_timer_key - initialize a timer
595 * @timer: the timer to be initialized
596 * @name: name of the timer
597 * @key: lockdep class key of the fake lock used for tracking timer
598 * sync lock dependencies
600 * init_timer_key() must be done to a timer prior calling *any* of the
601 * other timer functions.
603 void init_timer_key(struct timer_list
*timer
,
605 struct lock_class_key
*key
)
608 __init_timer(timer
, name
, key
);
610 EXPORT_SYMBOL(init_timer_key
);
612 void init_timer_deferrable_key(struct timer_list
*timer
,
614 struct lock_class_key
*key
)
616 init_timer_key(timer
, name
, key
);
617 timer_set_deferrable(timer
);
619 EXPORT_SYMBOL(init_timer_deferrable_key
);
621 static inline void detach_timer(struct timer_list
*timer
,
624 struct list_head
*entry
= &timer
->entry
;
626 debug_deactivate(timer
);
628 __list_del(entry
->prev
, entry
->next
);
631 entry
->prev
= LIST_POISON2
;
635 * We are using hashed locking: holding per_cpu(tvec_bases).lock
636 * means that all timers which are tied to this base via timer->base are
637 * locked, and the base itself is locked too.
639 * So __run_timers/migrate_timers can safely modify all timers which could
640 * be found on ->tvX lists.
642 * When the timer's base is locked, and the timer removed from list, it is
643 * possible to set timer->base = NULL and drop the lock: the timer remains
646 static struct tvec_base
*lock_timer_base(struct timer_list
*timer
,
647 unsigned long *flags
)
648 __acquires(timer
->base
->lock
)
650 struct tvec_base
*base
;
653 struct tvec_base
*prelock_base
= timer
->base
;
654 base
= tbase_get_base(prelock_base
);
655 if (likely(base
!= NULL
)) {
656 spin_lock_irqsave(&base
->lock
, *flags
);
657 if (likely(prelock_base
== timer
->base
))
659 /* The timer has migrated to another CPU */
660 spin_unlock_irqrestore(&base
->lock
, *flags
);
667 __mod_timer(struct timer_list
*timer
, unsigned long expires
,
668 bool pending_only
, int pinned
)
670 struct tvec_base
*base
, *new_base
;
674 timer_stats_timer_set_start_info(timer
);
675 BUG_ON(!timer
->function
);
677 base
= lock_timer_base(timer
, &flags
);
679 if (timer_pending(timer
)) {
680 detach_timer(timer
, 0);
681 if (timer
->expires
== base
->next_timer
&&
682 !tbase_get_deferrable(timer
->base
))
683 base
->next_timer
= base
->timer_jiffies
;
690 debug_activate(timer
, expires
);
692 cpu
= smp_processor_id();
694 #if defined(CONFIG_NO_HZ) && defined(CONFIG_SMP)
695 if (!pinned
&& get_sysctl_timer_migration() && idle_cpu(cpu
)) {
696 int preferred_cpu
= get_nohz_load_balancer();
698 if (preferred_cpu
>= 0)
702 new_base
= per_cpu(tvec_bases
, cpu
);
704 if (base
!= new_base
) {
706 * We are trying to schedule the timer on the local CPU.
707 * However we can't change timer's base while it is running,
708 * otherwise del_timer_sync() can't detect that the timer's
709 * handler yet has not finished. This also guarantees that
710 * the timer is serialized wrt itself.
712 if (likely(base
->running_timer
!= timer
)) {
713 /* See the comment in lock_timer_base() */
714 timer_set_base(timer
, NULL
);
715 spin_unlock(&base
->lock
);
717 spin_lock(&base
->lock
);
718 timer_set_base(timer
, base
);
722 timer
->expires
= expires
;
723 if (time_before(timer
->expires
, base
->next_timer
) &&
724 !tbase_get_deferrable(timer
->base
))
725 base
->next_timer
= timer
->expires
;
726 internal_add_timer(base
, timer
);
729 spin_unlock_irqrestore(&base
->lock
, flags
);
735 * mod_timer_pending - modify a pending timer's timeout
736 * @timer: the pending timer to be modified
737 * @expires: new timeout in jiffies
739 * mod_timer_pending() is the same for pending timers as mod_timer(),
740 * but will not re-activate and modify already deleted timers.
742 * It is useful for unserialized use of timers.
744 int mod_timer_pending(struct timer_list
*timer
, unsigned long expires
)
746 return __mod_timer(timer
, expires
, true, TIMER_NOT_PINNED
);
748 EXPORT_SYMBOL(mod_timer_pending
);
751 * Decide where to put the timer while taking the slack into account
754 * 1) calculate the maximum (absolute) time
755 * 2) calculate the highest bit where the expires and new max are different
756 * 3) use this bit to make a mask
757 * 4) use the bitmask to round down the maximum time, so that all last
761 unsigned long apply_slack(struct timer_list
*timer
, unsigned long expires
)
763 unsigned long expires_limit
, mask
;
766 expires_limit
= expires
;
768 if (timer
->slack
>= 0) {
769 expires_limit
= expires
+ timer
->slack
;
771 unsigned long now
= jiffies
;
773 /* No slack, if already expired else auto slack 0.4% */
774 if (time_after(expires
, now
))
775 expires_limit
= expires
+ (expires
- now
)/256;
777 mask
= expires
^ expires_limit
;
781 bit
= find_last_bit(&mask
, BITS_PER_LONG
);
783 mask
= (1 << bit
) - 1;
785 expires_limit
= expires_limit
& ~(mask
);
787 return expires_limit
;
791 * mod_timer - modify a timer's timeout
792 * @timer: the timer to be modified
793 * @expires: new timeout in jiffies
795 * mod_timer() is a more efficient way to update the expire field of an
796 * active timer (if the timer is inactive it will be activated)
798 * mod_timer(timer, expires) is equivalent to:
800 * del_timer(timer); timer->expires = expires; add_timer(timer);
802 * Note that if there are multiple unserialized concurrent users of the
803 * same timer, then mod_timer() is the only safe way to modify the timeout,
804 * since add_timer() cannot modify an already running timer.
806 * The function returns whether it has modified a pending timer or not.
807 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
808 * active timer returns 1.)
810 int mod_timer(struct timer_list
*timer
, unsigned long expires
)
813 * This is a common optimization triggered by the
814 * networking code - if the timer is re-modified
815 * to be the same thing then just return:
817 if (timer_pending(timer
) && timer
->expires
== expires
)
820 expires
= apply_slack(timer
, expires
);
822 return __mod_timer(timer
, expires
, false, TIMER_NOT_PINNED
);
824 EXPORT_SYMBOL(mod_timer
);
827 * mod_timer_pinned - modify a timer's timeout
828 * @timer: the timer to be modified
829 * @expires: new timeout in jiffies
831 * mod_timer_pinned() is a way to update the expire field of an
832 * active timer (if the timer is inactive it will be activated)
833 * and not allow the timer to be migrated to a different CPU.
835 * mod_timer_pinned(timer, expires) is equivalent to:
837 * del_timer(timer); timer->expires = expires; add_timer(timer);
839 int mod_timer_pinned(struct timer_list
*timer
, unsigned long expires
)
841 if (timer
->expires
== expires
&& timer_pending(timer
))
844 return __mod_timer(timer
, expires
, false, TIMER_PINNED
);
846 EXPORT_SYMBOL(mod_timer_pinned
);
849 * add_timer - start a timer
850 * @timer: the timer to be added
852 * The kernel will do a ->function(->data) callback from the
853 * timer interrupt at the ->expires point in the future. The
854 * current time is 'jiffies'.
856 * The timer's ->expires, ->function (and if the handler uses it, ->data)
857 * fields must be set prior calling this function.
859 * Timers with an ->expires field in the past will be executed in the next
862 void add_timer(struct timer_list
*timer
)
864 BUG_ON(timer_pending(timer
));
865 mod_timer(timer
, timer
->expires
);
867 EXPORT_SYMBOL(add_timer
);
870 * add_timer_on - start a timer on a particular CPU
871 * @timer: the timer to be added
872 * @cpu: the CPU to start it on
874 * This is not very scalable on SMP. Double adds are not possible.
876 void add_timer_on(struct timer_list
*timer
, int cpu
)
878 struct tvec_base
*base
= per_cpu(tvec_bases
, cpu
);
881 timer_stats_timer_set_start_info(timer
);
882 BUG_ON(timer_pending(timer
) || !timer
->function
);
883 spin_lock_irqsave(&base
->lock
, flags
);
884 timer_set_base(timer
, base
);
885 debug_activate(timer
, timer
->expires
);
886 if (time_before(timer
->expires
, base
->next_timer
) &&
887 !tbase_get_deferrable(timer
->base
))
888 base
->next_timer
= timer
->expires
;
889 internal_add_timer(base
, timer
);
891 * Check whether the other CPU is idle and needs to be
892 * triggered to reevaluate the timer wheel when nohz is
893 * active. We are protected against the other CPU fiddling
894 * with the timer by holding the timer base lock. This also
895 * makes sure that a CPU on the way to idle can not evaluate
898 wake_up_idle_cpu(cpu
);
899 spin_unlock_irqrestore(&base
->lock
, flags
);
901 EXPORT_SYMBOL_GPL(add_timer_on
);
904 * del_timer - deactive a timer.
905 * @timer: the timer to be deactivated
907 * del_timer() deactivates a timer - this works on both active and inactive
910 * The function returns whether it has deactivated a pending timer or not.
911 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
912 * active timer returns 1.)
914 int del_timer(struct timer_list
*timer
)
916 struct tvec_base
*base
;
920 timer_stats_timer_clear_start_info(timer
);
921 if (timer_pending(timer
)) {
922 base
= lock_timer_base(timer
, &flags
);
923 if (timer_pending(timer
)) {
924 detach_timer(timer
, 1);
925 if (timer
->expires
== base
->next_timer
&&
926 !tbase_get_deferrable(timer
->base
))
927 base
->next_timer
= base
->timer_jiffies
;
930 spin_unlock_irqrestore(&base
->lock
, flags
);
935 EXPORT_SYMBOL(del_timer
);
939 * try_to_del_timer_sync - Try to deactivate a timer
940 * @timer: timer do del
942 * This function tries to deactivate a timer. Upon successful (ret >= 0)
943 * exit the timer is not queued and the handler is not running on any CPU.
945 * It must not be called from interrupt contexts.
947 int try_to_del_timer_sync(struct timer_list
*timer
)
949 struct tvec_base
*base
;
953 base
= lock_timer_base(timer
, &flags
);
955 if (base
->running_timer
== timer
)
958 timer_stats_timer_clear_start_info(timer
);
960 if (timer_pending(timer
)) {
961 detach_timer(timer
, 1);
962 if (timer
->expires
== base
->next_timer
&&
963 !tbase_get_deferrable(timer
->base
))
964 base
->next_timer
= base
->timer_jiffies
;
968 spin_unlock_irqrestore(&base
->lock
, flags
);
972 EXPORT_SYMBOL(try_to_del_timer_sync
);
975 * del_timer_sync - deactivate a timer and wait for the handler to finish.
976 * @timer: the timer to be deactivated
978 * This function only differs from del_timer() on SMP: besides deactivating
979 * the timer it also makes sure the handler has finished executing on other
982 * Synchronization rules: Callers must prevent restarting of the timer,
983 * otherwise this function is meaningless. It must not be called from
984 * interrupt contexts. The caller must not hold locks which would prevent
985 * completion of the timer's handler. The timer's handler must not call
986 * add_timer_on(). Upon exit the timer is not queued and the handler is
987 * not running on any CPU.
989 * The function returns whether it has deactivated a pending timer or not.
991 int del_timer_sync(struct timer_list
*timer
)
993 #ifdef CONFIG_LOCKDEP
996 local_irq_save(flags
);
997 lock_map_acquire(&timer
->lockdep_map
);
998 lock_map_release(&timer
->lockdep_map
);
999 local_irq_restore(flags
);
1003 int ret
= try_to_del_timer_sync(timer
);
1009 EXPORT_SYMBOL(del_timer_sync
);
1012 static int cascade(struct tvec_base
*base
, struct tvec
*tv
, int index
)
1014 /* cascade all the timers from tv up one level */
1015 struct timer_list
*timer
, *tmp
;
1016 struct list_head tv_list
;
1018 list_replace_init(tv
->vec
+ index
, &tv_list
);
1021 * We are removing _all_ timers from the list, so we
1022 * don't have to detach them individually.
1024 list_for_each_entry_safe(timer
, tmp
, &tv_list
, entry
) {
1025 BUG_ON(tbase_get_base(timer
->base
) != base
);
1026 internal_add_timer(base
, timer
);
1032 static void call_timer_fn(struct timer_list
*timer
, void (*fn
)(unsigned long),
1035 int preempt_count
= preempt_count();
1037 #ifdef CONFIG_LOCKDEP
1039 * It is permissible to free the timer from inside the
1040 * function that is called from it, this we need to take into
1041 * account for lockdep too. To avoid bogus "held lock freed"
1042 * warnings as well as problems when looking into
1043 * timer->lockdep_map, make a copy and use that here.
1045 struct lockdep_map lockdep_map
= timer
->lockdep_map
;
1048 * Couple the lock chain with the lock chain at
1049 * del_timer_sync() by acquiring the lock_map around the fn()
1050 * call here and in del_timer_sync().
1052 lock_map_acquire(&lockdep_map
);
1054 trace_timer_expire_entry(timer
);
1056 trace_timer_expire_exit(timer
);
1058 lock_map_release(&lockdep_map
);
1060 if (preempt_count
!= preempt_count()) {
1061 WARN_ONCE(1, "timer: %pF preempt leak: %08x -> %08x\n",
1062 fn
, preempt_count
, preempt_count());
1064 * Restore the preempt count. That gives us a decent
1065 * chance to survive and extract information. If the
1066 * callback kept a lock held, bad luck, but not worse
1067 * than the BUG() we had.
1069 preempt_count() = preempt_count
;
1073 #define INDEX(N) ((base->timer_jiffies >> (TVR_BITS + (N) * TVN_BITS)) & TVN_MASK)
1076 * __run_timers - run all expired timers (if any) on this CPU.
1077 * @base: the timer vector to be processed.
1079 * This function cascades all vectors and executes all expired timer
1082 static inline void __run_timers(struct tvec_base
*base
)
1084 struct timer_list
*timer
;
1086 spin_lock_irq(&base
->lock
);
1087 while (time_after_eq(jiffies
, base
->timer_jiffies
)) {
1088 struct list_head work_list
;
1089 struct list_head
*head
= &work_list
;
1090 int index
= base
->timer_jiffies
& TVR_MASK
;
1096 (!cascade(base
, &base
->tv2
, INDEX(0))) &&
1097 (!cascade(base
, &base
->tv3
, INDEX(1))) &&
1098 !cascade(base
, &base
->tv4
, INDEX(2)))
1099 cascade(base
, &base
->tv5
, INDEX(3));
1100 ++base
->timer_jiffies
;
1101 list_replace_init(base
->tv1
.vec
+ index
, &work_list
);
1102 while (!list_empty(head
)) {
1103 void (*fn
)(unsigned long);
1106 timer
= list_first_entry(head
, struct timer_list
,entry
);
1107 fn
= timer
->function
;
1110 timer_stats_account_timer(timer
);
1112 set_running_timer(base
, timer
);
1113 detach_timer(timer
, 1);
1115 spin_unlock_irq(&base
->lock
);
1116 call_timer_fn(timer
, fn
, data
);
1117 spin_lock_irq(&base
->lock
);
1120 set_running_timer(base
, NULL
);
1121 spin_unlock_irq(&base
->lock
);
1126 * Find out when the next timer event is due to happen. This
1127 * is used on S/390 to stop all activity when a CPU is idle.
1128 * This function needs to be called with interrupts disabled.
1130 static unsigned long __next_timer_interrupt(struct tvec_base
*base
)
1132 unsigned long timer_jiffies
= base
->timer_jiffies
;
1133 unsigned long expires
= timer_jiffies
+ NEXT_TIMER_MAX_DELTA
;
1134 int index
, slot
, array
, found
= 0;
1135 struct timer_list
*nte
;
1136 struct tvec
*varray
[4];
1138 /* Look for timer events in tv1. */
1139 index
= slot
= timer_jiffies
& TVR_MASK
;
1141 list_for_each_entry(nte
, base
->tv1
.vec
+ slot
, entry
) {
1142 if (tbase_get_deferrable(nte
->base
))
1146 expires
= nte
->expires
;
1147 /* Look at the cascade bucket(s)? */
1148 if (!index
|| slot
< index
)
1152 slot
= (slot
+ 1) & TVR_MASK
;
1153 } while (slot
!= index
);
1156 /* Calculate the next cascade event */
1158 timer_jiffies
+= TVR_SIZE
- index
;
1159 timer_jiffies
>>= TVR_BITS
;
1161 /* Check tv2-tv5. */
1162 varray
[0] = &base
->tv2
;
1163 varray
[1] = &base
->tv3
;
1164 varray
[2] = &base
->tv4
;
1165 varray
[3] = &base
->tv5
;
1167 for (array
= 0; array
< 4; array
++) {
1168 struct tvec
*varp
= varray
[array
];
1170 index
= slot
= timer_jiffies
& TVN_MASK
;
1172 list_for_each_entry(nte
, varp
->vec
+ slot
, entry
) {
1173 if (tbase_get_deferrable(nte
->base
))
1177 if (time_before(nte
->expires
, expires
))
1178 expires
= nte
->expires
;
1181 * Do we still search for the first timer or are
1182 * we looking up the cascade buckets ?
1185 /* Look at the cascade bucket(s)? */
1186 if (!index
|| slot
< index
)
1190 slot
= (slot
+ 1) & TVN_MASK
;
1191 } while (slot
!= index
);
1194 timer_jiffies
+= TVN_SIZE
- index
;
1195 timer_jiffies
>>= TVN_BITS
;
1201 * Check, if the next hrtimer event is before the next timer wheel
1204 static unsigned long cmp_next_hrtimer_event(unsigned long now
,
1205 unsigned long expires
)
1207 ktime_t hr_delta
= hrtimer_get_next_event();
1208 struct timespec tsdelta
;
1209 unsigned long delta
;
1211 if (hr_delta
.tv64
== KTIME_MAX
)
1215 * Expired timer available, let it expire in the next tick
1217 if (hr_delta
.tv64
<= 0)
1220 tsdelta
= ktime_to_timespec(hr_delta
);
1221 delta
= timespec_to_jiffies(&tsdelta
);
1224 * Limit the delta to the max value, which is checked in
1225 * tick_nohz_stop_sched_tick():
1227 if (delta
> NEXT_TIMER_MAX_DELTA
)
1228 delta
= NEXT_TIMER_MAX_DELTA
;
1231 * Take rounding errors in to account and make sure, that it
1232 * expires in the next tick. Otherwise we go into an endless
1233 * ping pong due to tick_nohz_stop_sched_tick() retriggering
1239 if (time_before(now
, expires
))
1245 * get_next_timer_interrupt - return the jiffy of the next pending timer
1246 * @now: current time (in jiffies)
1248 unsigned long get_next_timer_interrupt(unsigned long now
)
1250 struct tvec_base
*base
= __get_cpu_var(tvec_bases
);
1251 unsigned long expires
;
1253 spin_lock(&base
->lock
);
1254 if (time_before_eq(base
->next_timer
, base
->timer_jiffies
))
1255 base
->next_timer
= __next_timer_interrupt(base
);
1256 expires
= base
->next_timer
;
1257 spin_unlock(&base
->lock
);
1259 if (time_before_eq(expires
, now
))
1262 return cmp_next_hrtimer_event(now
, expires
);
1267 * Called from the timer interrupt handler to charge one tick to the current
1268 * process. user_tick is 1 if the tick is user time, 0 for system.
1270 void update_process_times(int user_tick
)
1272 struct task_struct
*p
= current
;
1273 int cpu
= smp_processor_id();
1275 /* Note: this timer irq context must be accounted for as well. */
1276 account_process_tick(p
, user_tick
);
1278 rcu_check_callbacks(cpu
, user_tick
);
1280 perf_event_do_pending();
1282 run_posix_cpu_timers(p
);
1286 * This function runs timers and the timer-tq in bottom half context.
1288 static void run_timer_softirq(struct softirq_action
*h
)
1290 struct tvec_base
*base
= __get_cpu_var(tvec_bases
);
1292 hrtimer_run_pending();
1294 if (time_after_eq(jiffies
, base
->timer_jiffies
))
1299 * Called by the local, per-CPU timer interrupt on SMP.
1301 void run_local_timers(void)
1303 hrtimer_run_queues();
1304 raise_softirq(TIMER_SOFTIRQ
);
1309 * The 64-bit jiffies value is not atomic - you MUST NOT read it
1310 * without sampling the sequence number in xtime_lock.
1311 * jiffies is defined in the linker script...
1314 void do_timer(unsigned long ticks
)
1316 jiffies_64
+= ticks
;
1321 #ifdef __ARCH_WANT_SYS_ALARM
1324 * For backwards compatibility? This can be done in libc so Alpha
1325 * and all newer ports shouldn't need it.
1327 SYSCALL_DEFINE1(alarm
, unsigned int, seconds
)
1329 return alarm_setitimer(seconds
);
1337 * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this
1338 * should be moved into arch/i386 instead?
1342 * sys_getpid - return the thread group id of the current process
1344 * Note, despite the name, this returns the tgid not the pid. The tgid and
1345 * the pid are identical unless CLONE_THREAD was specified on clone() in
1346 * which case the tgid is the same in all threads of the same group.
1348 * This is SMP safe as current->tgid does not change.
1350 SYSCALL_DEFINE0(getpid
)
1352 return task_tgid_vnr(current
);
1356 * Accessing ->real_parent is not SMP-safe, it could
1357 * change from under us. However, we can use a stale
1358 * value of ->real_parent under rcu_read_lock(), see
1359 * release_task()->call_rcu(delayed_put_task_struct).
1361 SYSCALL_DEFINE0(getppid
)
1366 pid
= task_tgid_vnr(current
->real_parent
);
1372 SYSCALL_DEFINE0(getuid
)
1374 /* Only we change this so SMP safe */
1375 return current_uid();
1378 SYSCALL_DEFINE0(geteuid
)
1380 /* Only we change this so SMP safe */
1381 return current_euid();
1384 SYSCALL_DEFINE0(getgid
)
1386 /* Only we change this so SMP safe */
1387 return current_gid();
1390 SYSCALL_DEFINE0(getegid
)
1392 /* Only we change this so SMP safe */
1393 return current_egid();
1398 static void process_timeout(unsigned long __data
)
1400 wake_up_process((struct task_struct
*)__data
);
1404 * schedule_timeout - sleep until timeout
1405 * @timeout: timeout value in jiffies
1407 * Make the current task sleep until @timeout jiffies have
1408 * elapsed. The routine will return immediately unless
1409 * the current task state has been set (see set_current_state()).
1411 * You can set the task state as follows -
1413 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1414 * pass before the routine returns. The routine will return 0
1416 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1417 * delivered to the current task. In this case the remaining time
1418 * in jiffies will be returned, or 0 if the timer expired in time
1420 * The current task state is guaranteed to be TASK_RUNNING when this
1423 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1424 * the CPU away without a bound on the timeout. In this case the return
1425 * value will be %MAX_SCHEDULE_TIMEOUT.
1427 * In all cases the return value is guaranteed to be non-negative.
1429 signed long __sched
schedule_timeout(signed long timeout
)
1431 struct timer_list timer
;
1432 unsigned long expire
;
1436 case MAX_SCHEDULE_TIMEOUT
:
1438 * These two special cases are useful to be comfortable
1439 * in the caller. Nothing more. We could take
1440 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1441 * but I' d like to return a valid offset (>=0) to allow
1442 * the caller to do everything it want with the retval.
1448 * Another bit of PARANOID. Note that the retval will be
1449 * 0 since no piece of kernel is supposed to do a check
1450 * for a negative retval of schedule_timeout() (since it
1451 * should never happens anyway). You just have the printk()
1452 * that will tell you if something is gone wrong and where.
1455 printk(KERN_ERR
"schedule_timeout: wrong timeout "
1456 "value %lx\n", timeout
);
1458 current
->state
= TASK_RUNNING
;
1463 expire
= timeout
+ jiffies
;
1465 setup_timer_on_stack(&timer
, process_timeout
, (unsigned long)current
);
1466 __mod_timer(&timer
, expire
, false, TIMER_NOT_PINNED
);
1468 del_singleshot_timer_sync(&timer
);
1470 /* Remove the timer from the object tracker */
1471 destroy_timer_on_stack(&timer
);
1473 timeout
= expire
- jiffies
;
1476 return timeout
< 0 ? 0 : timeout
;
1478 EXPORT_SYMBOL(schedule_timeout
);
1481 * We can use __set_current_state() here because schedule_timeout() calls
1482 * schedule() unconditionally.
1484 signed long __sched
schedule_timeout_interruptible(signed long timeout
)
1486 __set_current_state(TASK_INTERRUPTIBLE
);
1487 return schedule_timeout(timeout
);
1489 EXPORT_SYMBOL(schedule_timeout_interruptible
);
1491 signed long __sched
schedule_timeout_killable(signed long timeout
)
1493 __set_current_state(TASK_KILLABLE
);
1494 return schedule_timeout(timeout
);
1496 EXPORT_SYMBOL(schedule_timeout_killable
);
1498 signed long __sched
schedule_timeout_uninterruptible(signed long timeout
)
1500 __set_current_state(TASK_UNINTERRUPTIBLE
);
1501 return schedule_timeout(timeout
);
1503 EXPORT_SYMBOL(schedule_timeout_uninterruptible
);
1505 /* Thread ID - the internal kernel "pid" */
1506 SYSCALL_DEFINE0(gettid
)
1508 return task_pid_vnr(current
);
1512 * do_sysinfo - fill in sysinfo struct
1513 * @info: pointer to buffer to fill
1515 int do_sysinfo(struct sysinfo
*info
)
1517 unsigned long mem_total
, sav_total
;
1518 unsigned int mem_unit
, bitcount
;
1521 memset(info
, 0, sizeof(struct sysinfo
));
1524 monotonic_to_bootbased(&tp
);
1525 info
->uptime
= tp
.tv_sec
+ (tp
.tv_nsec
? 1 : 0);
1527 get_avenrun(info
->loads
, 0, SI_LOAD_SHIFT
- FSHIFT
);
1529 info
->procs
= nr_threads
;
1535 * If the sum of all the available memory (i.e. ram + swap)
1536 * is less than can be stored in a 32 bit unsigned long then
1537 * we can be binary compatible with 2.2.x kernels. If not,
1538 * well, in that case 2.2.x was broken anyways...
1540 * -Erik Andersen <andersee@debian.org>
1543 mem_total
= info
->totalram
+ info
->totalswap
;
1544 if (mem_total
< info
->totalram
|| mem_total
< info
->totalswap
)
1547 mem_unit
= info
->mem_unit
;
1548 while (mem_unit
> 1) {
1551 sav_total
= mem_total
;
1553 if (mem_total
< sav_total
)
1558 * If mem_total did not overflow, multiply all memory values by
1559 * info->mem_unit and set it to 1. This leaves things compatible
1560 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1565 info
->totalram
<<= bitcount
;
1566 info
->freeram
<<= bitcount
;
1567 info
->sharedram
<<= bitcount
;
1568 info
->bufferram
<<= bitcount
;
1569 info
->totalswap
<<= bitcount
;
1570 info
->freeswap
<<= bitcount
;
1571 info
->totalhigh
<<= bitcount
;
1572 info
->freehigh
<<= bitcount
;
1578 SYSCALL_DEFINE1(sysinfo
, struct sysinfo __user
*, info
)
1584 if (copy_to_user(info
, &val
, sizeof(struct sysinfo
)))
1590 static int __cpuinit
init_timers_cpu(int cpu
)
1593 struct tvec_base
*base
;
1594 static char __cpuinitdata tvec_base_done
[NR_CPUS
];
1596 if (!tvec_base_done
[cpu
]) {
1597 static char boot_done
;
1601 * The APs use this path later in boot
1603 base
= kmalloc_node(sizeof(*base
),
1604 GFP_KERNEL
| __GFP_ZERO
,
1609 /* Make sure that tvec_base is 2 byte aligned */
1610 if (tbase_get_deferrable(base
)) {
1615 per_cpu(tvec_bases
, cpu
) = base
;
1618 * This is for the boot CPU - we use compile-time
1619 * static initialisation because per-cpu memory isn't
1620 * ready yet and because the memory allocators are not
1621 * initialised either.
1624 base
= &boot_tvec_bases
;
1626 tvec_base_done
[cpu
] = 1;
1628 base
= per_cpu(tvec_bases
, cpu
);
1631 spin_lock_init(&base
->lock
);
1633 for (j
= 0; j
< TVN_SIZE
; j
++) {
1634 INIT_LIST_HEAD(base
->tv5
.vec
+ j
);
1635 INIT_LIST_HEAD(base
->tv4
.vec
+ j
);
1636 INIT_LIST_HEAD(base
->tv3
.vec
+ j
);
1637 INIT_LIST_HEAD(base
->tv2
.vec
+ j
);
1639 for (j
= 0; j
< TVR_SIZE
; j
++)
1640 INIT_LIST_HEAD(base
->tv1
.vec
+ j
);
1642 base
->timer_jiffies
= jiffies
;
1643 base
->next_timer
= base
->timer_jiffies
;
1647 #ifdef CONFIG_HOTPLUG_CPU
1648 static void migrate_timer_list(struct tvec_base
*new_base
, struct list_head
*head
)
1650 struct timer_list
*timer
;
1652 while (!list_empty(head
)) {
1653 timer
= list_first_entry(head
, struct timer_list
, entry
);
1654 detach_timer(timer
, 0);
1655 timer_set_base(timer
, new_base
);
1656 if (time_before(timer
->expires
, new_base
->next_timer
) &&
1657 !tbase_get_deferrable(timer
->base
))
1658 new_base
->next_timer
= timer
->expires
;
1659 internal_add_timer(new_base
, timer
);
1663 static void __cpuinit
migrate_timers(int cpu
)
1665 struct tvec_base
*old_base
;
1666 struct tvec_base
*new_base
;
1669 BUG_ON(cpu_online(cpu
));
1670 old_base
= per_cpu(tvec_bases
, cpu
);
1671 new_base
= get_cpu_var(tvec_bases
);
1673 * The caller is globally serialized and nobody else
1674 * takes two locks at once, deadlock is not possible.
1676 spin_lock_irq(&new_base
->lock
);
1677 spin_lock_nested(&old_base
->lock
, SINGLE_DEPTH_NESTING
);
1679 BUG_ON(old_base
->running_timer
);
1681 for (i
= 0; i
< TVR_SIZE
; i
++)
1682 migrate_timer_list(new_base
, old_base
->tv1
.vec
+ i
);
1683 for (i
= 0; i
< TVN_SIZE
; i
++) {
1684 migrate_timer_list(new_base
, old_base
->tv2
.vec
+ i
);
1685 migrate_timer_list(new_base
, old_base
->tv3
.vec
+ i
);
1686 migrate_timer_list(new_base
, old_base
->tv4
.vec
+ i
);
1687 migrate_timer_list(new_base
, old_base
->tv5
.vec
+ i
);
1690 spin_unlock(&old_base
->lock
);
1691 spin_unlock_irq(&new_base
->lock
);
1692 put_cpu_var(tvec_bases
);
1694 #endif /* CONFIG_HOTPLUG_CPU */
1696 static int __cpuinit
timer_cpu_notify(struct notifier_block
*self
,
1697 unsigned long action
, void *hcpu
)
1699 long cpu
= (long)hcpu
;
1703 case CPU_UP_PREPARE
:
1704 case CPU_UP_PREPARE_FROZEN
:
1705 err
= init_timers_cpu(cpu
);
1707 return notifier_from_errno(err
);
1709 #ifdef CONFIG_HOTPLUG_CPU
1711 case CPU_DEAD_FROZEN
:
1712 migrate_timers(cpu
);
1721 static struct notifier_block __cpuinitdata timers_nb
= {
1722 .notifier_call
= timer_cpu_notify
,
1726 void __init
init_timers(void)
1728 int err
= timer_cpu_notify(&timers_nb
, (unsigned long)CPU_UP_PREPARE
,
1729 (void *)(long)smp_processor_id());
1733 BUG_ON(err
!= NOTIFY_OK
);
1734 register_cpu_notifier(&timers_nb
);
1735 open_softirq(TIMER_SOFTIRQ
, run_timer_softirq
);
1739 * msleep - sleep safely even with waitqueue interruptions
1740 * @msecs: Time in milliseconds to sleep for
1742 void msleep(unsigned int msecs
)
1744 unsigned long timeout
= msecs_to_jiffies(msecs
) + 1;
1747 timeout
= schedule_timeout_uninterruptible(timeout
);
1750 EXPORT_SYMBOL(msleep
);
1753 * msleep_interruptible - sleep waiting for signals
1754 * @msecs: Time in milliseconds to sleep for
1756 unsigned long msleep_interruptible(unsigned int msecs
)
1758 unsigned long timeout
= msecs_to_jiffies(msecs
) + 1;
1760 while (timeout
&& !signal_pending(current
))
1761 timeout
= schedule_timeout_interruptible(timeout
);
1762 return jiffies_to_msecs(timeout
);
1765 EXPORT_SYMBOL(msleep_interruptible
);